Blog

Perspectives on entrepreneurship, startups and venture capital from K9 Ventures.

Gradescope: AI for Grading and Assessment

Tweet about this on TwitterShare on FacebookShare on LinkedInShare on Google+Email this to someone

 

“I love grading” — I doubt those words have ever been uttered in earnest! The only times I’ve ever heard them or even used them is with a healthy dose of sarcasm. Lot’s of people (me included) will say “I love teaching,” but that love for grading just doesn’t come naturally. Grading gets tedious quickly due to its repetitive nature. Engineers in general hate doing things that are repetitive. If it has to be repeated then let’s build a machine to do it more efficiently!

 

logo_teal_transparent

 

Well, that’s exactly what the founders of Gradescope thought. Arjun Singh and Sergey Karayev were both PhD students in CS at UC Berkeley. They co-founded Gradescope with Pieter Abbeel, a Stanford PhD and an Associate Professor of Robotics and Machine Learning at UC Berkeley, and Ibrahim Awwal. Berkeley has fairly large class sizes and as TAs Arjun and Sergey had a heavy load of grading to do. To make their work easier they started building some tools to help them spend less time grading. Fast forward and today they have over 100,000 students that have been graded on Gradescope, with over 8M questions graded to date. They’re being used by all the top schools in CS/STEM including Berkeley, Stanford, Harvard, MIT, Carnegie Mellon, Michigan, UW and over 100 more.

 

 

The mission of Gradescope is nothing short of revolutionizing the education system by eventually eliminating letter grades and replacing them with concept-based assessment. My theory on letter grades is that they originated as the simplest way for a instructor to bucket a student’s progress, with one letter, one byte of information being transferred over. However, in this day and age, it makes no sense to say someone got a A or B, instead we should know what concepts they understand and what concepts they need to work on more. My best illustration of this came from a comedian who said, “My friend just passed his pilot’s exam. He said he scored an 89. How do I know the 11 points he missed weren’t about landing!?”

 

Animated Product

 

Gradescope has already built the tools that help instructors and TAs to spend about half the time doing grading compared to without using their tools. They use computer vision to segment students answers. Graders can be co-located or remote (something that was hard to do before!) and each graders can be assigned questions, so that they can quickly and efficiently run through the questions while being in a state of flow. At the same time Gradescope allows them to maintain a rubric for assigning grades. Let’s say after grading 20+ students exams, I decide that taking of 2 points for an error is too much and taking of 1 point would be better. Previously, graders would never even consider that since it would mean they have to manually go through and adjust all the 20 exams they’ve already graded. With Gradescope, it’s automatic and takes no time at all!

Students receive feedback directly on Gradescope, so they don’t only get to see what they scored, but also get to see why and in the process actually learn what they missed on the test/assignment.

But that’s not all. The Gradescope team are PhDs in AI and machine learning. They’re already applying their expertise towards making the problem of grading even simpler yet. By automatically clustering all the similar answers in a test, a grader now needs to grade far fewer responses than they would have to before. Imagine all the correct responses being automatically scored based on learning the actions performed by the grader. Likewise, all the incorrect answers can be grouped into common errors and be graded in chunks.

I don’t want to steal all their thunder in this post, so you’ll just have to wait for some of the other cool stuff that the Gradescope team is already working on. It suffices to say that they’re the first and the only team I’ve encountered who is applying Computer Vision, AI and Machine Learning to the problem of grading and by doing that they’re building the platform where current and future students will get highly personalized feedback on their learning and performance. Gradescope is bringing big data to education.
K9 led the Pre-Seed round for Gradescope in October 2014, and I’ve enjoyed working with Arjun and Sergey to take Gradescope to the next level. Today the company announced that it had raised $2.6M in a Seed round led by Freestyle Capital, with K9, Bloomberg Beta, and Reach Capital participating in the round. I’m pleased to welcome Dave Samuel from Freestyle Capital as a fellow board member and looking forward to the future for Gradescope. Congrats Arjun, Sergey, Pieter, Ibrahim and the extended Gradescope Team!

 

Gradescope Founders
If you’re a professor or a TA, you should absolutely check out Gradescope and see if they can help save you time, effort, and aggravation for grading your homework assignments and tests.

If you’re a techie looking to join a startup, Gradescope is hiring.

You can follow me on Twitter at @ManuKumar or @K9Ventures for just the K9 Ventures related tweets. K9 Ventures is also on Facebook and Google+.